Google Inc.

Olga Kapralova

Gleb Skobeltsyn

Evgeny A Cherepanov

Martin Baeuml

Dmitry Osmakov

Pedro Jose Moreno Mengibar

Ignacio Lopez Moreno

Vladimir Vuskovic

Bradley Nelson

Eugene Weinstein

Olivier Siohan

Abraham Lee

Mikhail Reutov

Yufei Zhao

John Paul Alex

Konstantin Shagin

Dan Vallejo

Wendy Look

Alessio Macrí

Alessio Macrì

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for speech recognition. One of the methods includes receiving first audio data corresponding to an utterance; obtaining a first transcription of the first audio data; receiving data indicating (i) a selection of one or more terms of the first transcription and (ii) one or more of replacement terms; determining that one or more of the replacement terms are classified as a correction of one or more of the selected terms; in response to determining that the one or more of the replacement terms are classified as a correction of the one or more of the selected terms, obtaining a first portion of the first audio data that corresponds to one or more terms of the first transcription; and using the first portion of the first audio data that is associated with the one or more terms of the first transcription to train an acoustic model for recognizing the one or more of the replacement terms.

Acoustic model training using corrected terms

Implementations set forth herein relate to an automated assistant that can provide interactive application widgets based on their relevance to content that a user may have expressed interest in. The automated assistant can render the application widgets according to an estimated familiarity of the user with the content they expressed interest in. Each application widget can correspond to an application that can be accessed separately from the automated assistant. An application widget can be rendered at a display interface simultaneous to the user accessing the content that served as the basis for rendering the application widget. When the user interacts with the application widget, the automated assistant can communicate selection data to a corresponding application, which can respond with supplemental data that can be rendered at the display interface.

Interactive application widgets rendered with assistant content

Implementations set forth herein relate to incorporating automated assistant suggestions into an interface of a video application, when the video application is rendering video content. The video content—as well as any relevant content, can provide a basis for the automated assistant suggestions. The assistant suggestions can optionally link to one or more additional applications, which can be controlled in response to a user selecting one or more of the automated assistant suggestions. In response to a selection of an assistant suggestion, resulting data generated by another application can be rendered over an interface of the video application, while video content is being rendered and/or otherwise played. This can allow the user to control relevant actions of other applications without completely leaving an interface of the video application, thereby preserving memory and other computational resources that may be consumed when switching between application interfaces.

Generating and/or causing rendering of video playback-based assistant suggestion(s) that link to other application(s)

Implementations described herein relate to providing suggestions, via a display modality, for completing a spoken utterance for an automated assistant, in order to reduce a frequency and/or a length of time that the user will participate in a current and/or subsequent dialog session with the automated assistant. A user request can be compiled from content of an ongoing spoken utterance and content of any selected suggestion elements. When a currently compiled portion of the user request (from content of a selected suggestion(s) and an incomplete spoken utterance) is capable of being performed via the automated assistant, any actions corresponding to the currently compiled portion of the user request can be performed via the automated assistant. Furthermore, any further content resulting from performance of the actions, along with any discernible context, can be used for providing further suggestions.

Supplementing voice inputs to an automated assistant according to selected suggestions

The present disclosure relates to training a speech recognition system. One example method includes receiving a collection of speech data items, wherein each speech data item corresponds to an utterance that was previously submitted for transcription by a production speech recognizer. The production speech recognizer uses initial production speech recognizer components in generating transcriptions of speech data items. A transcription for each speech data item is generated using an offline speech recognizer, and the offline speech recognizer components are configured to improve speech recognition accuracy in comparison with the initial production speech recognizer components. The updated production speech recognizer components are trained for the production speech recognizer using a selected subset of the transcriptions of the speech data items generated by the offline speech recognizer. An updated production speech recognizer component is provided to the production speech recognizer for use in transcribing subsequently received speech data items.

Acoustic model training corpus selection

Growing community of inventors

Bern, Switzerland

Olga Kapralova